Manoj Bajaj — AI Agent Infrastructure

I build the infrastructure layer between "agents can do things" and "we can safely let agents do things."

Co-founder of Ruzo.ai. Building authsome.ai — an open-source credential broker for AI agents — and writing about agent-native infrastructure at for-agents.com.

Authsome handles OAuth2 and API-key injection for AI agents without exposing credentials to the subprocess environment. Agents can be built without solving identity, but cannot be safely deployed in production.

Before this: six years in B2B enterprise software, proprietary trading infrastructure at Squarepoint Capital, and early-stage at FalconX. IIT Delhi, mechanical engineering, 2017.

I write here about production systems, AI agent primitives, and the unglamorous infrastructure work that makes autonomous systems safe to operate.

Writing

May 28, 2026

You’re Still Building for the Wrong User

May 28, 2026

The Agent Economy Needs a New Stack

May 14, 2026

Agent Tooling Should Be Composable, Not Convenient

April 17, 2026

Agent Identity: Core agent primitive

March 24, 2026

Minimal Agent Primitives

March 12, 2026

MCP vs CLI is wrong question

March 6, 2026

Don't Build Agents, Build Primitives.